github/ColossalAI
1
0
Fork 0
mirror of https://github.com/hpcaitech/ColossalAI.git synced 2025-09-07 20:10:17 +00:00
Code Issues Packages Projects Releases Wiki Activity

[async io]supoort async io (#6137)

Browse Source 
* support async optimizer save/load

* fix

* fix

* support pin mem

* Update low_level_zero_plugin.py

* fix

* fix

* fix

* fix

* fix
This commit is contained in:
flybird11111
2024-11-18 17:52:24 +08:00
committed by Hongxin Liu
parent b90835bd32
commit eb69e640e5
15 changed files with 374 additions and 46 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
colossalai/checkpoint_io/hybrid_parallel_checkpoint_io.py
View File

@@ -416,6 +416,7 @@ class HybridParallelCheckpointIO(GeneralCheckpointIO):
gather_dtensor: bool = True,
prefix: Optional[str] = None,
size_per_shard: int = 1024,
use_async: bool = False,
):
"""
Save sharded optimizer checkpoint under the given checkpointing path.
@@ -725,7 +726,9 @@ class HybridParallelCheckpointIO(GeneralCheckpointIO):
# Update master params if mixed-precision training is enabled.
model_before_wrapping.update_master_params()
def save_unsharded_optimizer(self, optimizer: OptimizerWrapper, checkpoint: str, gather_dtensor: bool):
def save_unsharded_optimizer(
self, optimizer: OptimizerWrapper, checkpoint: str, gather_dtensor: bool, use_async: bool = False
):
"""
Save optimizer state dict to a file with given path.